WPS 16-03-2 Working Paper Series

Property Rights and Land Disputes: Theory and Evidence from Ethiopia

Salvatore Di Falco

Jérémy Laurent-Lucchetti

Marcella Veronesi

Gunnar Kohlin

March 2016

Property Rights and Land Disputes: Theory and

Evidence from Ethiopia

Salvatore Di Falco∗ Jeremy Laurent-Lucchetti†

Marcella Veronesi‡ Gunnar Kohlin§

March 23, 2016

Abstract

This paper investigates the link between insecure property rights and

land disputes using farm-household panel data from Ethiopia. We offer

two main contributions. First, we develop a novel theoretical framework

of land disputes. Our model predicts that in difficult times (i.e. when

water is scarce), bargaining is more likely to breakdown—and dispute to

arise— if property rights are ill-defined. Second, guided by our theo-

retical framework, we empirically assess the causal relationship between

land tenure security and clashes. Our identification strategy relies on the

gradual rollout of a land certification program at the village level and on

the exogenous variation in water availability, a trigger of land disputes.

We find that having tenure security reduces the likelihood for a farm-

household to experience land disputes by about 40%. We also show that

∗University of Geneva, Geneva School of Economics and Management.†University of Geneva, Geneva School of Economics and Management.‡University of Verona, Department of Economics; and ETH Zurich, Center for Development

and Cooperation (NADEL).§University of Gothenburg.

1

secure property rights are more likely to reduce land disputes when house-

holds face adverse weather shocks, hence by reducing the vulnerability to

water scarcity. We further document that water scarcity has a stronger

impact on land disputes when the marginal value of land is larger.

JEL classification: D74, O13.

Keywords: Property rights, land disputes, water scarcity, bargaining, Ethiopia,

tenure security.

2

1 Introduction

Insecure property rights over land are an important factor behind social conflicts

and violence in the developing world. Case studies linking ill-defined property

rights and land disputes abound in the literature: the fact that many individu-

als may have claims to the same piece of land is alleged to exacerbate tensions

and generate mistrust, especially in times of scarcity.1 Furthermore, it is docu-

mented that in periods of insecurity land related disputes can turn increasingly

violent and are linked to broader security, livelihood, political and identity is-

sues.2 The key challenge lies in identifying the causal effect of tenure security on

land disputes.

While there exists a growing body of literature studying the causal channel

linking insecure property rights to various economic outcomes—such as educa-

tion, access to credit or migration3—little attention has been paid to the mech-

anisms linking tenure insecurity and land disputes. This is surprising because

in many developing countries, informal institutions—socially shared, unwritten

rules—are the main way communities manage land rights (Blattman et al., 2014).

It is often pointed out that the lack of central enforcement of these informal in-

stitutions is a reason for deadlocks, disagreements and even violence (Fearon,

1998). Furthermore, it is usually purported that these bargaining failures are

more likely to occur when agents face adverse shocks: it may be more difficult

to sustain cooperation in times of scarcity (Ostrom, 2005).

In this paper, we offer two main contributions. First, we develop a novel

theoretical framework of land disputes. Our simple framework predicts that in

1A typical illustration is the Kenyan case where, in the aftermath of the early 2008 post-election violence, the ”Kenya National Dialogue and Reconciliation” process identified landreform as key to peace and reconciliation.

2See for example the report on ”Land and Conflict” produced by the ”Disaster and conflict”program of the UNEP: United Nations Environment Programme (2011).

3See for recent examples Besley (1995), Goldstein and Udry (2008), Besley et al. (2012),Acemoglu et al. (2014), or de Janvry et al. (2015).

3

difficult times (i.e. when water is scarce), bargaining is more likely to break

down—and dispute to arise— if property rights are ill-defined. Second, guided

by our theoretical framework, we empirically assess the causal relationship be-

tween tenure security and land disputes. We provide new evidence showing that

households with tenure security are less prone to land disputes.

Following the theoretical framework, we also detail a mechanism linking prop-

erty rights and tenure security. In particular, we show that secure property rights

are (i) more likely to reduce land disputes when households face adverse weather

shocks and (ii) that water scarcity has a stronger impact on land disputes when

the marginal value of land is larger. Identifying a clear mechanism linking tenure

security, water scarcity and land disputes is new in the literature and is of impor-

tance to effectively reduce the occurrence of land disputes in developing countries.

Our theoretical framework is a simple bargaining model with private infor-

mation. We assume that agents bargain over a piece of land. One agent can

illegally use the land, albeit at a (privately known) cost.4 This cost reflects the

strength of property rights in the region: the cost is substantive if tenure se-

curity is strong, and low if property rights are ill-defined. When facing water

scarcity, the temptation to encroach on the land of the tenant rises along with

the marginal value of land. This setup gives rise to a simple perfect Bayesian

equilibrium in which we observe (i) fewer costly disagreements when property

rights are more secure and (ii) higher probability of disagreement in dire times.

The framework highlights that the marginal value of land is central in explaining

the role of tenure security and water scarcity on land disputes.

We empirically test the theoretical framework by using data on the gradual

4Most rational models of conflicts (see Blattman and Miguel, 2010 for a recent review) arebased on a contest mechanism: agents invest in a fighting technology determining a probabilityof winning the fight. We are interested in a different type of events, encroachment and illegaluse of land, in which physical violence is almost never used directly. The cost of disputes isnot linked directly to the investment in a conflict technology but rather reflects the legal andreputation costs of encroaching.

4

rollout at the village level of a large land certification program implemented by

the World Bank in the Highlands of Ethiopia. In particular, we use a large

farm-household panel survey conducted in the region in years 2005 and 2007.

Our identification strategy relies on a fixed effects specification that compares

reported land disputes between households in certified and non-certified villages.

Because the program provided certificates to the entire village simultaneously,

this process eliminates concerns about selection at the individual level. Any

time-invariant village characteristic correlated with the program rollout and land

disputes is accounted for by the village fixed effects. Therefore, the identifying

assumption for our first result—the effect of tenure security on land disputes—is

that any time-varying village characteristic that affects land disputes is uncor-

related with the rollout of the program. We provide evidences supporting this

assumption in sections 4 and 5.

Our second main result—the effect of tenure security on land dispute transits

through less vulnerability to water scarcity—relies on exogenous variations of

water availability. We use a drought indicator, the Standardized Precipitation-

Evapotranspiration Index (SPEI), reflecting the climatic water balance at dif-

ferent time scales. This metric allows to investigate whether farm-households

with tenure security are more resilient to weather anomalies, and less likely to

experience disputes.

We first show that having tenure security reduces the likelihood for a house-

hold to experience land disputes by about 40%. We then present robust evidence

that water scarcity affects disputes over land: our baseline estimates suggest that

switching from a good year in which no serious droughts occur during the rainy

season to a bad year in which severe droughts occur every month during the

rainy season increases the likelihood of disputes over land by around 24% for

farm-households without land tenure. For households with tenure security this

effect is not significantly different from 0. This highlights that tenure security

5

leads to fewer disputes by reducing the vulnerability of rural households to water

scarcity. Guided by the predictions of our theoretical model, we further docu-

ment that factors increasing the marginal value of land (i.e. having more labor or

livestock available in the farm) magnify the impact of water scarcity on land dis-

putes. Our findings are robust to different specifications and measures of climate

shocks.

Ethiopia provides a prime set up to address our research questions. A largely

rural country, historically plagued by rainfall anomalies, Ethiopia’s land rights

were shaped by a radical land reform in 1975 implemented by the Marxist regime

(the Derg). As a result, all land was made collective property of the Ethiopian

people for more than three decades. Farmers were given usufruct rights (only

user-rights to land) with periodic village-level land redistributions. Land could

not be sold, mortgaged, exchanged, or transferred. While the new government

in 1991 partially lifted the market ban and then allowed for some land rental

activities, tenure security remained a critical issue. The momentum for more

market oriented land policy was indeed very weak. A combination of oral con-

tracts and lack of judicial institutions to intervene in case of disputes created

a situation of persistent land insecurity, and frequent litigations among rural

households (Deininger et al., 2008). The land certification program issued non-

alienable use-right certificates and used a participatory and highly decentralized

process of field adjudication.

This paper is related to a vast empirical literature on how improved prop-

erty rights affect economic outcomes. In reviewing the property rights literature,

Besley and Ghatak (2009) detail how well-defined and secure property rights over

land can impact economic outcomes. First, tenure security can increase invest-

ment incentives (Besley, 1995; Fenske, 2011) and it can also increase the use of

land as collateral in accessing credit (Besley et al., 2012). Furthermore, there is

evidence that political power uses land tenure as a way to control the local econ-

6

omy. For instance, Goldstein and Udry (2008) show that in Ghana agents who

are not central to the networks of local political power through which land is allo-

cated are very likely to have their land expropriated if it is fallowed. This creates

an incentive for intensive and inefficient land use. Similarly, in a recent contri-

bution, Acemoglu et al. (2014) show that in Sierra Leone powerful chiefs control

the access to the land. As a result, a whole series of development outcomes,

such as educational attainment and child health among others, are significantly

lower. Finally, De Janvry et al. (2015), using the rollout of the Mexican land

certification program from 1993 to 2006, show that land certification induces

migration.

In this study, we document a new channel though which tenure security im-

pacts economic outcomes: insecure property rights lead to disputes over land in

dire times. Land security is crucial for many development outcomes given that

localized disputes can severely affect welfare, generate tensions and potentially

escalate into large-scale conflicts (Andre and Platteau, 1998).

Our results also add new micro evidence to the literature on the relation-

ship between weather anomalies and conflicts.5 It is generally found that rainfall

anomalies are positively correlated with the occurrence of conflicts. We show

here, using micro data, that tenure security is an important channel through

which water scarcity translates into local clashes. It is noteworthy that this

literature stresses an ”opportunity cost” mechanism: water scarcity generates

violence through a decrease in agricultural income, which diminishes the oppor-

tunity cost of fighting for farmers (Chassang and Padro-i-Miquel, 2009). In this

study, we deal with disputes over land and encroachment: no physical fight is

usually taking place. We highlight that for this kind of disputes the marginal

value of land is at the roots of the mechanism linking water scarcity and land

disputes (and not income). To the best of our knowledge, the clear identification

5See for instance, recent studies by Jia (2013); Harari and La Ferrara (2014); Couttenierand Soubeyran (2014) or Almer et al. (2015).

7

of the role of tenure insecurity in generating disputes over land is new in this

literature.

The paper is organized as follows. Section 2 presents the theoretical frame-

work. Section 3 describes the background and the Ethiopian land certification

program. Section 4 describes the data while section 5 presents the empirical

strategy and the results. Section 6 investigates the underlying mechanisms link-

ing property rights and land disputes and section 7 provides concluding remarks.

2 Theoretical Framework

This section presents a simple theoretical framework close in spirit to Besley and

Ghatak (2009). Taking stock of the fact that physical violence is almost never

used directly for land disputes, we rely on a bargaining model with asymmetric

information: if bargaining breaks down one agent simply encroaches on the land

of the other. Our model highlights the role of the marginal value of land in

generating land disputes in dire times.

The environment Two agents N = {1, 2} share a total amount of land of

size 1. Let x1 be the land share of agent 1 and x2 the share of agent 2, with

x1 + x2 = 1. We assume that water falls uniformly over the land. We denote by

w the amount of water available per unit of land. Hence, each agent has access

to an amount of water xiw which we denote wi.

Water is a substantial input in production (i.e. farming, herding) but also

an essential part of agents livelihoods. We denote the payoff that each agent

gets from an amount wi of water by bi(wi − w) where w > 0 represents the

minimum water requirement of an individual. We assume that bi(.) is increasing

and concave in the available amount of water wi and that ∂bi(.)∂wi

goes to infinity

as wi − w tends to 0. Let vi be the marginal value of land for each agent, i.e.

∂bi∂xi

= vi for i = 1, 2.

8

Agents also value money. The utility of agent i from consuming wi units of

water and receiving a net money transfer ti is ui = bi(wi − w) + ti. We assume

that agents are expected-utility maximizers.

The game The agents bargain over a piece of land owned by agent 1 (i.e. a

part of x1). Bargaining occurs naturally because the marginal value of land of

agent 1 (”the tenant”) is lower than agent 2, i.e. v2 > v1. The game proceeds as

follows: agent 1 makes an offer p to agent 2 for the piece of land. Agent 2 can

either accept or refuse this offer. If agent 2 refuses the offer, he can either use the

land illegally or start a (Nash) bargaining procedure.6 An important simplifying

feature is that disputing or bargaining are game ending moves (as in most of

the literature, see Alesina and Spolaore, 2005 or Chassang and Padro-i-Miquel,

2010). This implies that once agent 2 starts bargaining the choice of encroaching

is forgone (i.e. the bargaining happens in front of the local authority—the elder

of the village in our empirical study—and the outcome is enforceable). The

bargaining procedure will lead agents to agree on a price p for the land such that

v2 > p > v1. At this price the gains in utility from the transfer of land will be

(p− v1) for agent 1 and (v2 − p) for agent 2.7

If agent 2 contests the negotiation and decides to encroach on the tenant’s

land she has to incur a cost c.8 This cost reflects the legal formalities associ-

ated with the illegal use of land and the social cost of potentially acquiring bad

reputation for encroaching, for example. We assume that this cost is uniformly

distributed9 over [θ − 12, θ + 1

2] and is private information of agent 2. The pa-

6Note that the outcome of the game is qualitatively similar in the symmetric situation whereagent 2 first makes an offer, which can be accepted or refused by agent 1, in which case agent2 can either encroach over the land or start a Nash bargaining procedure.

7We choose the Nash bargaining over a non-cooperative procedure for simplicity. However,it is known that the Nash bargaining solution is a good approximation of the non-cooperativeoutcome of a two-person bargaining game a la Rubinstein (1982) when time delay goes to 0(Binmore, 1987).

8This cost is sunk and is the source of inefficiency in our framework.9We assume a uniform distribution for the sake of transparency. However, as it will become

9

rameter θ summarizes the strength of property rights: a higher θ reflects that

property rights are more secure in the region.

2.1 Analysis

We proceed by backward induction, first studying the outcome of the bargaining

procedure.

The bargaining procedure We are looking for the Nash solution of the bar-

gaining process, hence the price for the land p maximizing the Nash product

Maxp(p− v1)α1(v2 − p)α2 (1)

where α2 = 1− α1 is exogenously given and summarize the bargaining power of

agent 1 and agent 2 (respectively).

The first order condition leads to α1(p − v1)α1−1 = α2(v2 − p)α2−1. Dividing

both sides by (p− v1)α1−1(v2 − p)α2−1 and rearranging leads to

p∗ = (1− α1)v1 + α1v2 (2)

Note that v1 ≤ p∗ ≤ v2.

The decision of agent 2: dispute or peaceful negotiation Agent 2 com-

pares the benefits of the initial offer and the bargaining price with the cost of

seizing the land illegally:

- If min{p, p∗, c} = p agent 2 accepts.

- If min{p, p∗, c} = c, agent 2 encroaches on the piece of land.

- Finally, if min{p, p∗, c} = p∗ agent 2 refuses the initial offer and starts

bargaining.

clear while detailing the equilibrium, most of our results hold with more general distributionfunctions, in particular symmetric distributions with (weakly) decreasing hazard rates.

10

The offer of agent 1 The decision of agent 1 depends on θ and on the sub-

sequent decisions highlighted above. Agent 1 first seeks the p maximizing her

expected payoff: (P{c ≥ p}) (p − v1) − (P{c < p}) v1, leading to p = θ2

+ 14

with

our distributional assumption. We denote this solution by pmax:

- If pmax ≤ p∗, agent 1 offers p = pmax. Any other initial offer decreases

the expected payoff of agent 1 given that pmax maximizes the expected payoff of

agent 1.

- If pmax > p∗, agent 1 offers p = p∗. If p > p∗, agent 2 refuses the ini-

tial offer and starts the bargaining procedure anyway which leads to p∗ (if not

encroaching). Any offers p < p∗ lead to a lower payoff for agent 1.

2.2 Equilibrium

In a perfect Bayesian equilibrium of the game, the strategies of the agents are as

follows: Agent 1’s strategy is to

- propose p = pmax if pmax ≤ p∗;

- propose p = p∗ if pmax > p∗.

Agent 2’s strategy is to

- accept any initial offer if min{p, p∗, c} = p;

- reject the initial offer and encroach if min{p, p∗, c} = c;

- reject the initial offer and start the bargaining process if min{p, p∗, c} = p∗.

Proposition 1 In a perfect Bayesian Nash equilibrium of the game, agent 1

offers p = min{pmax, p∗}. Agent 2 accepts the initial offer if p ≤ c and encroaches

if p > c.

2.3 From Theory to Evidence

This simple framework can be used to obtain comparative statics predictions

resulting from household and water availability heterogeneity. We show that

11

the probability of observing a dispute varies with the strength of the property

rights and further depends on water scarcity, farm size and factors impacting the

marginal value of land.

• Prediction 1. We should observe that more secure property rights dimin-

ish land disputes. More secure property rights are reflected as a higher θ:

better property rights translate in a higher average value for c, the cost of

encroaching. Disputes occur at the equilibrium when p = min{pmax, p∗} >

c. When p = p∗, clashes are less likely when θ is higher, simply because

P{c ≤ p∗} ≡ p∗ − θ + 12

decreases with θ. When p = pmax, the probability

of disputes is also decreasing with θ because P{c ≤ pmax} ≡ 1− θ2.

• Prediction 2. A decrease in the amount of water available to agents

should increase the probability of land disputes through an increase of

the marginal value of land. As shown above, the probability of clashes is

increasing in p∗ when p = p∗, because P{c ≤ p∗} ≡ p∗ − θ + 12. It is clear

in equation (2) that an increase in v1 and/or v2 leads to an increase in p∗.

Recalling that vi is decreasing in wi (bi(.) is concave), a decrease in the

amount of water wi toward the minimal level w will increase both v1 and

v2, and ultimately p∗.

• Prediction 3. As a corollary, any factor increasing v1 and/or v2 will also

increase the price p∗ and the probability of disputes when p = p∗.

• Prediction 4. An increase in x1—the land share of agent 1— may increase

the probability of disputes. Recall that x1 = 1− x2, hence a higher x1 will

decrease v1 because ∂v1∂x1≤ 0 (recall that bi(.) is concave). Symmetrically,

v2 will increase through a decrease of x2. Hence, referring to equation (2)

we can say that p∗ rises after an increase in x1 if

12

dp∗

dx1

= α1dv2

dx1

+ (1− α1)dv1

dx1

≥ 0. (3)

Hence, if

dv2

dx1

≥ −1− α1

α1

dv1

dx1

(4)

Assuming α1 = α2, this implies that land inequality increases the probabil-

ity of disputes if the marginal value of land is convex, i.e. if ∂v2∂x2≤ ∂v1

∂x1≤ 0

when v1 < v2.10 If agent 2 has a very high bargaining power, land inequal-

ity may actually decrease the probability of disputes (because the price will

already reflect more closely the interests of agent 2).

3 Background and the Ethiopian Land Certifi-

cation Program

3.1 Background on the Ethiopian Land Tenure System

The land tenure system of Ethiopia was characterized by private land holding

(the gult system), communal land tenure (the rist system) or the church until

year 1975. The gult system was the relevant system in the southern half of

Ethiopia and was characterized by absentee owners while the rist system was the

dominant system in the northern half of the country and was denoted by shared

land rights and land distribution based on the principle of equality (Bezabih et

al., 2011). In the rist system, first the land was divided into parcels according to

quality, and then, a lottery was implemented to allocate the land. In addition, if

claimants could establish a direct line of descendants from the recognized original

10This assumption is very standard in demand analysis (see Deaton, 1992 for an example)and is met by many standard functions such as the Cobb-Douglas, the log funtion, the CRRAand the CARA utility function.

13

land holder then individuals obtained usufruct rights on the land. Under this

system, farmers could not transfer the land rights to others through mortgage or

sale.

In 1975 the empirical regime fell and a radical land reform was implemented.

Proclamation No. 31/1975 established the collective property of land (Nickola,

1988). Almost all subsistence producers received usufruct rights. However, land

transfer rights through bequests, leases, mortgages or sales, as well as hiring of

labor were prohibited. In year 1991 the Ethiopian Peoples Revolutionary Demo-

cratic Front replaced the Derg regime in power since 1975 and established the

state ownership of all land through Article 40 of the Constitution. Pastoralists

and peasants were guaranteed the rights to access the land, and to transfer, re-

move or claim compensation for any improvements they made on the land. In

addition, the ban on land market activity was removed giving incentives to the

development of a land rental market.

3.2 The Ethiopian Land Certification Program

The Ethiopian land certification program has delivered the largest number of

non-freehold land rights per time unit in Sub-Saharan Africa and follows the

1997 Federal Rural Land Administration Proclamation, which was revised in

2005. The program has been implemented in the four most populated regions

of Ethiopia: Tigray in years 1998 and 1999, Amhara in year 2003, Oromiya and

the Southern Nations, Nationalities and Peoples’ Region (SNNPR) in year 2004

(Deininger et al., 2008; Bezabih et al., 2011).

The focus of our analysis is the Amhara National Regional State (ANRS),

which is the second largest region in Ethiopia, located in the Northwest of the

country (see Figure A1 of the appendix for a map of the Amhara region).11

11The Amhara region covers 11% of Ethiopia’s total area and is divided into three majoragricultural climatic zones: lowland (below 1,500 meters), semi-highland (between 1,500 and2,300 meters) and highland (above 2,300 meters), accounting for 28%, 44% and 20% of the

14

Following the land proclamation at a federal level, ANRS adopted a Regional

Conservation Strategy (RCS) in year 1999, and a land administration and land

use policy in year 2000 to stop land degradation. Afterwards, proclamation No.

46/2000 (revised in year 2006) was issued. This proclamation established state

ownership of the land and the registration of farmland for lifetime entitlement

and children’s succession. In year 2000, ANRS established the regional Envi-

ronmental Protection, Land Administration and Use Authority (EPLAUA) to

implement the RCS and the land use policy (proclamation No. 47/2000).

The lowest level of the local government (kebele and sub-kebele) handled

the actual implementation process. Kebele and sub-kebele land administration

committees were elected by the local community and trained as land registrars. A

local consultation process took place before registration. The fees were generally

very low and the technology was very simple. This was a highly participatory

process with most of the input for adjudication and the demarcation of land

provided by the local community, including women.

The preparation and awareness campaign started with kebele and sub-kebele

administration discussions with the farmers, followed by the election of the Land

Administration and Use Committees (LAC) and the provision of training. This

was followed by the LAC and representatives of elders from the different kebe-

les together with the different kebele officials of EPLAUA demarcating kebeles’

boundaries, public areas and communal lands. Each potential certificate recip-

ient was instructed to negotiate the border with their neighbours. The LAC

together with the farmers first conducted the demarcation of parcel boundaries

and the traditional measurement of the area, and then after entering the infor-

mation into field sheets, the LAC approved the legal status of the holding. The

result of the land adjudication was presented to the public for a month long ver-

land area, respectively. A large part of the population is living in highland areas with steepslope topography and about 90% (14.7 million) of the people live in rural areas (Adenew andAbdi, 2005.

15

ification and corrections. The LAC chairperson and the kebele EPLAUA head

would then approve the field sheets, with outstanding conflicts being passed on to

the courts. Information collected regarding holdings and parcels were registered

in the Land Registry Book at the kebele and approved by the kebele EPLAUA

head and LAC chairperson.

The actual evidence of registration was issued in two stages. First, the LAC

of the kebele issued temporary certificates based on the approved field sheet

information, mainly as a way of ensuring that paper evidence was given to the

certificate recipients before the actual certificate. Then, the Book of Holding was

issued by the kebele (Palm, 2010). Types of data collected included: name and

address of the landholder; parcel boundaries of the landholder; the estimated

area of parcels; the current land use and land cover (grazing, cropland, etc.); the

level of fertility status (low, medium and poor) identified by subjective judgment;

and the local area name for identifying the parcels’ locations.

The major advantage of the certification program is a decentralized imple-

mentation process at the village level through the elected LAC that involves

sufficient public awareness about the program and the election of the local land

administration committee with little political interference (Palm, 2010; Deininger

et al., 2011). In addition, although land remains state-owned and many restric-

tions on land transfers continue to exist, certification was reported to increase

the chance of getting compensation in the case where land was acquired for non-

agricultural uses. Deininger et al. (2011) find that the land certification program

in Ethiopia increased investments related to the land, renting and resulted highly

cost effective as well as improved female household heads’ participation in the

land rental market. Bezabih et al. (2011) analyze the effect of land certification

in Ethiopia on trust and productivity of female farmers, respectively, and find

significant impacts of the program in increasing trust and the productivity of

land owned by female-headed households.

16

An important feature of the certification program is that lack of manpower

led to adoption of a gradual rollout of the program across kebeles. Because the

program provided certificates to the entire village simultaneously, this process

eliminates concerns about selection at the individual level.12 As per discussions

with officials from the regional EPLAUA, the choice of the certification timing

was based on a combination of fixed characteristics of kebeles such as the kebele’s

administrative capacity and the kebele’s facilities.

Differences between certified and non-certified kebeles do not undermine our

identification strategy as long as they are uncorrelated with land disputes. To

address this concern we verify in sections 4 and 5 that the level of land disputes

in year 2005—prior to the program—is not correlated with the likelihood for a

household to be certified in year 2007. We also include many time-varying house-

hold characteristics as well as district-year fixed effects in our main analysis to

account for the possibility that land disputes changed over time due to unobserv-

able characteristics that were correlated with the timing of land certification.

4 Data

4.1 Land Disputes

We use the Sustainable Land Management Survey to estimate the effect of land

tenure certification on land disputes. This is a farm-household panel survey

conducted by the Department of Economics of Addis Ababa University in col-

laboration with the Ethiopian Development Research Institute, the University of

Gothenburg and the World Bank in 2005 and 2007.13 The survey was conducted

in the Amhara National Regional State of Ethiopia and includes 12 villages, six

12This stands in contrast to many titling program where allocation is demand driven.13There were actually four waves of this survey, conducted in 1999, 2002, 2005 and 2007.

However, only the last two waves have explicit information on land disputes, our outcome ofinterest.

17

from the East Gojjam zone and six from the South Wello zone. These villages

are part of seven districts (woredas). About 1,500 households were selected by

random stratified sampling based on indicators such as population density, access

to the market and agricultural potential.14

One of the survey instruments was specifically designed to elicit land disputes.

Specific questions were included to investigate whether farmers have experienced

any disputes: ”Have you ever faced any conflicts or claims regarding the land

you own ?” An event was defined as a conflict in the following cases: (i) the

claimant pushed the borders of the farmers parcel; (ii) it was claimed that the

plot was unfairly given to the farmer; (iii) it was claimed that the plot belonged

to the claimant sometime ago; (iv) it was claimed that the farmer pushed the

claimant’s borders; (v) the claimant did not want to leave the land the farmer

left for him to manage while the farmer was away; (vi) the claimant did not want

to leave the farmer’s land he had given out to him on sharecropping.

Our final sample includes 1,487 farm-households (1,027 with land tenure and

460 without land tenure) for a total of 2,974 observations (2,054 with land tenure

and 920 without land tenure). For the villages covered by the survey, the actual

provision of certificates to the farmers started in May 2005. By year 2007, four

villages have received the land certification (the ”certified” group) while eight

villages have not received the land certification yet (the ”non-certified” group).15

The four certified villages are found in three districts (Machakel, Tenta and Debre

Elias) and the eight non-certified villages are part of four other districts (Gozmin,

Enemay, Tehuldere and Harbu/Kalu). It is noteworthy that the first round of

the survey covered the period before the certification started (beginning of year

2005) and that all farm-households in certified villages had been certified before

14The rate of attrition was extremely low, less than 1% per year. It is thus highly unlikelythat sample selection due to attrition drives our main results.

15In the original sample, six villages are certified. However, two villages were used in a pretestand were thus already certified in 2005. We excludes these two villages from our analysis.

18

the implementation of the second round of the survey in year 2007.

Tables A1 and A2 of the appendix present descriptive statistics on many

certified and non-certified households characteristics at the village and household

level, respectively. Table A1 shows that certified and non-certified villages are

not statistically different in many characteristics at the baseline in year 2005

before the land certification was implemented, while Table A2 shows that at the

baseline households that will receive the certification are (i) more literate, (ii)

have more livestock, (iii) are more likely to use improved seed, (iv) less likely

to use soil conservation measures and (v) have less access to credit on average

than households that will not receive the certification.16 We will account for

the heterogeneity in these observable characteristics—potentially impacting land

disputes—in our main analysis.

We also observe that at the baseline in year 2005 our key variable of interest,

the indicator of reported land disputes, is not statistically different between cer-

tified and non-certified farm-households at the village and household level. We

ran a set of Kolmogorov-Smirnov tests for the comparison of the distributions of

land disputes in year 2005 in any possible pair of certified and non-certified vil-

lages and we never reject the null hypothesis of equality of the distributions out

of 34 tests (the lowest value being 0.31): certified villages have not significantly

different levels of land disputes than non-certified villages in year 2005 before

the implementation of the land certification. Tables A1 and A2 show that in

year 2007—after the program implementation—the proportion of land disputes

is significantly higher in the non-certified group than in the certified group.

Given that households have slightly different characteristics in certified and

non-certified groups, Tables A3 and A4 of the appendix report average treatment

effects of a matching estimator (propensity score matching) applied to reported

land disputes in years 2005 and 2007. We match households in certified and

16These results are all confirmed by Kolmogorov-Smirnov tests comparing the distributionsof the observable characteristics, at the 10% level.

19

non-certified groups on observable characteristics and we observe that in year

2005 there is no significant difference in land disputes, both in the unmatched

and the matched group.17 In year 2007, the non-certified group experiences a

significantly higher level of land disputes: 8.7% in the unmatched group and

10.3% in the matched sample.18 Interestingly, this effect is in line with the result

from our main analysis, where we account for different sources of heterogeneity by

including various sets of fixed effects and time-varying observable characteristics.

One important concern for our analysis is whether certified and non-certified

villages follow different time trends on land disputes due to unobserved hetero-

geneity (i.e. certified villages have a higher rate of decrease in land disputes than

non-certified villages due to better administrative capacities or other unobserv-

able characteristics). Given that we have data on land disputes only for two

time periods (before and after the certification), we are unable to provide direct

evidence for a common trend between certified and non-certified villages in re-

ported land disputes long before the implementation of the program (i.e. before

year 2005). However, our empirical analysis provides three sets of results sup-

porting the absence of time-trending unobserved heterogeneity correlated with

land disputes.

First, we use a proxy for reported land disputes present in past surveys on

the same sample of households. As mentioned, there was four waves of the

survey, conducted in years 1999, 2002, 2005 and 2007 and only the last two waves

had information on land disputes. However, in the four surveys we observe the

reported perception of tenure insecurity. Each household had to mention whether

a change in the land holding was expected in the next five years.19 It is likely

17We obtain the best balance by matching groups based on the farm size, the age and literacylevel of the head of the household, the size of the household, the livestock size, if the householdhas access to credit and if the farm-household uses improved seeds, soil conservation measuresor irrigation.

18These results are robust to many different balance specifications and matching estimators.Results available upon request.

19We observe this variable over the four surveys only for a subset of households: respec-

20

that perceived tenure insecurity and land disputes are positively correlated over

this period: a farmer may fear losing some land due to ongoing disputes with

neighbors or, conversely, the expectation of land reallocation may create tensions

among neighbors and generate disputes.

We observe in Figure A2 of the appendix that both certified and non-certified

kebeles report very similar perceived tenure insecurity from 1999 to 2005. Pos-

sibly as a result of the 1997 redistribution, perceived tenure insecurity in year

1999 was very high, with 78% and 75% of households expecting a change of land

holdings in certified and non-certified villages, respectively. In year 2005, before

the implementation of the land certification program, this proportion decreased

to 38% in certified and non-certified villages. In year 2007, plausibly because of

the certification the program, the trends start to diverge, dropping to 24% in cer-

tified villages while staying around 39% in non-certified villages. If one believes

that the perceived tenure insecurity in a village is correlated with land disputes

over time, Figure A2 provides a convincing argument for a common trend in land

disputes for certified and non-certified villages before the certification program.20

Another concern is whether more peaceful villages were selected for being

certified first due to unobservable characteristics. Our data show that the corre-

lation between a dummy variable indicating if a household was certified or not in

2007 and a dummy variable indicating whether a household experienced any land

disputes at the baseline is statistically not significant (-0.037; s.e. 0.106). This

provide additional evidence that land certification was not implemented based

on the propensity of being peaceful or not. In addition, our fixed effect analysis

rules out any selection bias due to time-invariant unobservable characteristics

tively 462 and 233 households for non-certified and certified kebeles in 1999, and 477 and 356households in non-certified and certified kebeles in 2007.

20While establishing our main result, one of our specifications will include explicitly thechange in land area between 2005 and 2007 for certified and non-certified households. Thisspecification will show that land reallocation is not a confounding factor driving the effect ofthe certification program on land disputes.

21

such as the propensity to be confrontational.

Finally, we will show in section (5) that our main result is invariant to the

inclusion of (i) village fixed effects; (ii) household fixed effects; (iii) time-varying

household characteristics and (iv) district-year fixed effects. The first three ro-

bustness checks account for the possibility that land disputes changed over time

due to characteristics that were (possibly) correlated with the timing of land

certification. The last robustness check aims to control for the possibility that

the program started earlier in certain types of districts that experienced differ-

ential changes in land disputes due to reasons other than land certification (i.e.

disputes decrease faster in a district because it has better administrative capac-

ities). Allowing for the time effects to depend on district fixed effects we further

control for district characteristics that might have changed overtime and might

have been correlated with the timing of the certification and the likelihood of

conflicts.

4.2 Climatic Data

Our main explanatory variable of water scarcity is the Standardized Precipitation-

Evapotranspiration Index (SPEI), developed by Vicente-Serrano et al. (2010).

SPEI is a drought index that reflects the climatic water balance at different time

scales and computes the difference between precipitation and potential evapo-

transpiration for each kebele over a specific time period (e.g., monthly or yearly).

SPEI is a standardized variable that expresses the water balance in units of stan-

dard deviations from the long-run average of a cell. A value of zero means that

the water balance is exactly at its long-run average; while a value of plus one (mi-

nus one) means that the water balance is one standard deviation above (below)

the long-run average.

We used two datasets for computing the SPEI index at the kebele level (see

appendix for more details). The precipitation data come from the African Rain-

22

fall Climatology Version 2 dataset (ARC2, Novella and Thiaw, 2013), which

provides daily estimates at a resolution of 0.1 decimal degree, from 1983 to 2014,

based on a combination of gauge and satellite data.21 The dataset has notably

been developed as a key input of the Famine Early Warning System Network

(FEWSNET), one of the main tool used by international humanitarian agencies

to monitor food security. The temperature data come from the Climate Pre-

diction Center Global Land Surface Air Temperature Analysis (GHCN+CAM,

Fan and Van den Dool, 2008) as monthly mean surface air temperatures at a 0.5

decimal degree resolution over the period 1948-2014. We provide more details

on the construction of our main climate variables in the appendix.

5 Empirical Strategy and Results

We now turn on the empirical investigation of the theoretical predictions pre-

sented in section 2. We will first investigate whether land certification decreases

the likelihood of land disputes (prediction 1) and whether farm-households with

secure property rights are less likely to experience conflicts triggered by adverse

weather shocks (prediction 2).

5.1 Land Certification and Land Disputes

Our empirical strategy on the effect of land certification on land disputes relies

on (i) the gradual rollout of the land certification program to farm-households

at the village level; and (ii) the panel nature of our dataset. In particular,

we estimate a fixed effects model where the identification of the effect of tenure

security on land disputes is coming from comparing reported disputes in year 2005

(before the implementation of the land certification) and in year 2007 (after the

21Given that we are dealing with local land disputes—and not large scale conflicts—we arenot worried about the fact that rain-gauges located at the ground level may be damaged bylocal fighting activities, a potential concern in the climate and conflict literature.

23

implementation of the land certification) in certified and non-certified villages.

We estimate the following equation:

Yijt = β0 + β1Ljt + µj + µt + ε1,ijt (5)

where Y is an indicator for whether a household i in village j reported a land

dispute in year t (t = 2005, 2007), L is a dummy variable equal to one if a house-

hold has secure property rights over land, µj and µt represent village (kebele)

and time fixed effects and ε is an idiosyncratic error term. Standard errors are

clustered at the village level. The inclusion of village fixed effects allows us to

account for any time-invariant village characteristics that are correlated with the

program rollout.

Table 1 presents the baseline results on the relationship between land certifi-

cation and land disputes. Fixed effects estimates of equation (5) assess the first

prediction of our theoretical model, that is whether tenure security reduces the

likelihood of land disputes. Column 1 presents results from the baseline specifi-

cation including village and year fixed effects. It shows that the probability of

a household experiencing land disputes decreases by 0.08 after being certified.

The average rate of land disputes during the sample period is 20.5%, indicating

that land certification decreases the likelihood of land disputes by 39%, a sizable

effect. This result is extremely robust to the inclusion of different set of fixed ef-

fects and clustering methods. Column 2 shows that the effect of land certification

remains unaffected when household fixed effects are included in the regression,

confirming that time-invariant household characteristics did not influence the

program rollout differently than time-invariant village characteristics.

One main concern is the possibility of differential time trends that would

be correlated with the timing of the certification and the likelihood of disputes.

Column 3 shows that the result is robust to controlling for district-year fixed

effects. The purpose of this robustness check is to control for the possibility that

24

the program was initiated earlier in certain types of districts that experienced

differential changes in land disputes due to reasons other than land certification.

Allowing for the time effects to depend on these fixed effects further controls for

district characteristics that (i) might have changed overtime and (ii) might have

been correlated with the timing of the certification and the likelihood of conflicts.

Finally, we account for different clustering methods, the results are presented

in Table A5 of the appendix. Given the small number of clusters (n < 13),

the standard errors could be biased downwards. In column 1, we therefore

re-estimate equation (5) by applying the finite sample correction suggested by

Cameron et al. (2013). Then, in column 2 we estimate two-way cluster-robust

standard errors to account for within-panel autocorrelation (clustering on farm-

households) and cross-panel correlation (clustering on time) following Cameron

et al. (2011). In column 3, we also re-estimate equation (5) by adjusting stan-

dard errors to account for potential spatial autocorrelation within one kilometer

(Hsiang, 2010).22 Results in Table A5 shows that the negative effect of land

certification on land disputes stays significant for all the specifications (p-value

< 0.065).

In addition, we re-estimate equation (5) accounting for time-varying char-

acteristics. Table 2 reports results where we control for a set of time-variant

farm-household characteristics. Column 1 addresses a potential concern of the

land certification program: land certification may have induced a land realloca-

tion impacting land disputes (through a reduction of tensions over land alloca-

tion, for example). We show that our result is only slightly lower in magnitude

when introducing a variable measuring the variation in surface of the farm size

between 2005 and 2007. Column 2 introduces other potential omitted variables

that change overtime such as the age and the literacy level of the household

22We re-estimated the model with different cutoffs—5, 10, 25 and 50 kilometers—and thecoefficient estimate of the land certification variable remains significant at the 10% level.

25

Table 1: Land Certification and Land Disputes: Baseline Results

(1) (2) (3)Land disputes Land disputes Land disputes

Land certification -0.080∗ -0.080∗ -0.080∗

(0.042) (0.042) (0.042)

Village fixed effects yes no yesHousehold fixed effects no yes noYear fixed effects yes yes yesDistrict × Year fixed effects no no yesObservations 2974 2974 2974R2 0.027 0.008 0.027

Notes: All regressions are linear probability models. The dependent variable is equal to 1 if the farm-household

has experienced any land disputes, 0 otherwise. Land certification is equal to 1 if the farm-household is certified,

0 otherwise. Standard errors clustered at the kebele level are reported in parentheses. ∗ Significant at the 10%

level.

head, the family size, the area of land cultivated, the total livestock owned by

the farm-household, if the farm-household uses improved seeds or not and, finally,

if soil conservation measures have been adopted on the farm. Column 3 further

includes interaction terms between time effects and each of these household-level

characteristics to account for specific time trends more flexibly. Once again our

main result is invariant to this richer specification: the sign, the magnitude and

the significance of the effect of tenure security on land disputes remain mostly

unchanged. This is consistent with the fact that the certificates were distributed

simultaneously to all households in a village. In summary, our analysis firmly

points to land certification decreasing the probability that a household experi-

ences land disputes.

5.2 Land Certification, Water Scarcity and Land Dis-

putes

We now turn to the mechanism linking tenure security and land disputes. The

objective is to investigate theoretical prediction 2, namely that (i) droughts in-

26

Table 2: Land Certification and Land Disputes: Robustness Checks

(1) (2) (3)Land disputes Land disputes Land disputes

Land certification -0.059∗∗∗ -0.083∗ -0.084∗∗

(0.003) (0.041) (0.035)

Change in farm size (2005-2007) 0.003(0.002)

Age of household head 0.000 0.001(0.001) (0.001)

Male household head 0.008 -0.047(0.021) (0.026)

Literacy of household head -0.005 0.021(0.017) (0.021)

Household size -0.003 0.012(0.016) (0.021)

Farm size -0.003 -0.008(0.022) (0.022)

Livestock 0.013 0.013(0.017) (0.017)

Using improved seeds 0.045∗∗ 0.030(0.018) (0.040)

Adopted soil conservation measures -0.024 0.004(0.028) (0.045)

Access to credit 0.014 0.005(0.022) (0.042)

Village fixed effects yes yes yesYear fixed effects yes yes yesHousehold characteristics × Year fixed effects no no yesObservations 1278 2799 2799R2 0.063 0.026 0.031

Notes: All regressions are linear probability models. The dependent variable is equal to 1 if the farm-household

has experienced any land disputes, 0 otherwise. Land certification is equal to 1 if the farm-household is certified,

0 otherwise. Literacy of the household head, household size, farm size and total livestock in the household are

dummy variables equal to 1 if greater than the median, 0 otherwise. The sample size is smaller in column 1

because of missing values in the reported farm size. Standard errors clustered at the kebele level are reported

in parentheses. ∗ Significant at 10% level; ∗∗ Significant at the 5% level; ∗∗∗ Significant at the 1% level.

27

crease the likelihood of land disputes (especially when they happen during the

rainy seasons); and (ii) farm-households with land certification are less prone to

land disputes when they face adverse weather shocks. We exploit the random na-

ture of weather shocks for the identification of the effects and introduce climate

conditions in equation (5). We compare the same household overtime subject to

different water availability during the main rainy season.

In particular, we investigate whether households with land certification are

less likely to experience land disputes triggered by water scarcity by estimating

Yijt = β0 + β1Ljt + β2Wijt + β5Wijt × Ljt + µj + µt + ε2,ijt (6)

where Y is the aforementioned land dispute dummy variable, L is the land cer-

tification dummy variable, W is a variable measuring water availability at the

farm-household location and W × L denotes the interaction of the water avail-

ability variable with the land certification dummy variable L. We report results

of equation (6) with different measures of W , accounting for the effect of water

scarcity during the two main rainy seasons in the region and for the effect of past

droughts on the reported level of land disputes (land disputes are reported in t

in the survey, but may have occurred in the past years).

We first provide estimates of equation (6) with different measures of water

scarcity, based on our computation of the drought index SPEI for each house-

hold. The SPEI index considers the combined effect of rainfall, potential evap-

otranspiration and temperature—a lower value of the index means less water

availability in the village. We also account for temporal lags of the index. Ta-

ble 3, column 1 shows that the signs of the parameter estimates for SPEI is

negative, as expected.23 Hence a negative water balance in the last year is asso-

23Note that the negative sign of the SPEI index coefficient indicates that less water duringthe rainy season increases the likelihood of disputes, and the positive sign in front of theinteraction term between SPEI and land certification means that tenure security reduces thiseffect.

28

ciated with a higher probability of land disputes. Furthermore we observe that

farm-households with land certification are far less prone to land disputes: the

positive coefficient in front of the interaction term counterbalance the previous

effect. Column 2 suggests that the effect of a negative water balance on land

conflicts is slightly stronger when focusing on the SPEI index of the last main

rainy season. As we can see from column 3, this effect increases again when

considering both rainy seasons (summer and spring). However, the tempering

effect of tenure security is less clear in columns 2 and 3 than in column 1.

Following the recent literature on water scarcity and conflicts24 we adopt a

finer instrument to measure water availability. Our benchmark indicator of water

scarcity, denoted as Strong Shock is defined as the fraction of the rainy seasons

(spring and summer) during which the SPEI index was below its mean by one

standard deviation (and between half a standard deviation and one standard

deviation for Mild Shock). Table 4 confirms previous results: (i) droughts during

both rainy seasons increase the likelihood of conflicts; and (ii) farm-households

with land certification are less prone to land disputes in difficult times. Columns

1-2 present results for shocks happening during the spring and the summer sea-

sons, column 3 presents results for both seasons separately and column 4 for the

seven consecutive months of both seasons. Column 4 shows that for mild shocks,

switching from the best possible year in terms of our climate variable to the worst

possible year (no month with a drought to every month of both rainy seasons

with a drought) increases the likelihood of observing land disputes by 24% for

households without land certification25. Interestingly, the effect of Mild Spei on

land disputes is not significantly different from zero if farm-households have been

certified. Hence, tenure security dampens disputes by reducing the vulnerability

24See for recent examples, Couttenier and Soubeyran (2014) or Harari and La Ferrara (2014).25The average farm-household without land certification faces a ”Mild Spei” value equal to

0.12 while the average rate of land disputes is equal to 0.22. This implies that for the averagefarm-household without certification the probability of experiencing land disputes increases by0.440.22 ∗ 0.12 = 24% after switching from the best possible year to the worst possible year.

29

Table 3: Land Certification, Water Scarcity and Land Disputes

(1) (2) (3)Land disputes Land disputes Land disputes

Land certification 0.094 -0.031 -0.062∗

(0.061) (0.034) (0.034)

SPEI (t− 1) -0.036∗∗

(0.012)

SPEI (t− 1) × Land certification 0.322∗∗∗

(0.058)

SPEI (t− 2) 0.006(0.010)

SPEI (t− 2) × Land certification 0.344∗∗∗

(0.075)

SPEI (summer season, t− 1) -0.040∗∗

(0.013)

SPEI (summer season, t− 1) × Land certification -0.063(0.066)

SPEI (summer season, t− 2) 0.042(0.034)

SPEI (summer season, t− 2) × Land certification 0.027(0.057)

SPEI (summer and spring seasons, t− 1) -0.045∗∗

(0.014)

SPEI (summer and spring seasons, t− 1) × Land certification -0.085(0.055)

SPEI (summer and spring seasons, t− 2) 0.011(0.020)

SPEI (summer and spring seasons, t− 2) × Land certification 0.040(0.036)

Village fixed effects yes yes yesYear fixed effects yes yes yesObservations 2734 2734 2734R2 0.025 0.025 0.025

Notes: All regressions are linear probability models. The dependent variable is equal to 1 if the farm-household

has experienced any land disputes, 0 otherwise. Land certification is equal to 1 if the farm-household is

certified, 0 otherwise. SPEI indicates water availability at the farm-household location and is defined in the

appendix. Standard errors clustered at the kebele level are reported in parentheses. ∗ Significant at 10% level;

∗∗ Significant at the 5% level; ∗∗∗ Significant at the 1% level.

30

of households to water scarcity: the effect of tenure security on disputes transits

mostly through a reduction in clashes triggered by water scarcity.

Finally, we investigate whether our specifications miss important aspects of

the data and we re-estimate the effect of SPEI shocks on the probability of land

disputes using a quadratic regression fit. Figure A3 of the appendix shows that

the relationship is monotonically decreasing: larger weather shocks are linked

with a higher probability of experiencing land disputes. Interestingly, tenure

security seems to reduce the probability of disputes for the whole range of SPEI

shocks.

5.3 Further Sensitivity Analysis

Results shown in Tables 1, 2 and 3 are robust to many sensitivity checks. We

present in the appendix additional robustness analysis. Table A6 reports results

based on non-linear specifications, namely Probit and Logit models. These esti-

mations are consistent with the results presented in the previous tables: tenure

security decreases the likelihood of land disputes. As expected, the effect is

stronger in quantitative terms when using non-linear specifications.

Table A7 presents results for different measures of land disputes. Column

1 shows the effect of tenure security on the onset of land disputes (a dummy

variable equal to one when a household reported no dispute in year 2005 and at

least one dispute in year 2007 and equal to zero otherwise). Certified households

experience significantly (at the 1% level) less emergence of disputes after the pro-

gram than non-certified households. Column 2 presents results on the decrease

of disputes (a dummy variable equal to one when a household reported a dispute

in year 2005 but none in year 2007 and equal to zero otherwise). We observe

that land certification significantly (at the 1% level) increases the likelihood that

a farm-household experiences a dispute in year 2005 and reported none in year

2007.

31

Table 4: Land Certification, Water Scarcity and Land Disputes: RobustnessChecks

(1) (2) (3) (4)Land disputes Land disputes Land disputes Land disputes

Land certification -0.047 -0.027 0.073 -0.008(0.071) (0.084) (0.052) (0.073)

Mild SPEI Shock (spring rainy season) 0.147∗∗ 0.213∗∗∗

(0.049) (0.053)

Mild SPEI Shock (spring rainy season)×Land certification -0.289∗ 0.048(0.141) (0.136)

Strong SPEI Shock (spring rainy season) -0.032 -0.068(0.065) (0.094)

Strong SPEI Shock (spring rainy season)×Land certification 0.006 -0.748∗∗∗

(0.203) (0.165)

Mild SPEI Shock (summer rainy season) 0.021 0.132∗

(0.087) (0.061)

Mild SPEI Shock (summer rainy season)×Land certification -0.184 -0.594∗∗∗

(0.207) (0.067)

Strong SPEI Shock (summer rainy season) -0.050 0.098∗

(0.100) (0.053)

Strong SPEI Shock (summer rainy season)×Land certification 0.059 -0.156(0.391) (0.132)

Strong SPEI Shock (spring and summer rainy seasons) -0.099(0.193)

Strong SPEI Shock (spring and summer rainy seasons) ×Land certification -0.456(0.533)

Mild SPEI Shock (spring and summer rainy seasons) 0.444∗∗∗

(0.138)

Mild SPEI Shock (spring and summer rainy seasons) ×Land certification -0.423∗∗

(0.180)

Village fixed effects yes yes yes yesYear fixed effects yes yes yes yesObservations 2734 2734 2734 2734R2 0.024 0.022 0.026 0.024

Notes: All regressions are linear probability models. The dependent variable is equal to 1 if the farm-household

has experienced any land disputes, 0 otherwise. Land certification is equal to 1 if the farm-household is

certified, 0 otherwise. Strong SPEI Shock is defined as the fraction of the rainy seasons (spring and summer)

during which the SPEI index was below its mean by one standard deviation (and between half a standard

deviation and one standard deviation for Mild SPEI Shock). More details on the SPEI index are provided in

the appendix. Standard errors clustered at the kebele level are reported in parentheses. ∗ Significant at 10%

level; ∗∗ Significant at the 5% level; ∗∗∗ Significant at the 1% level.

32

The baseline results are again confirmed when we estimate equation (6) with

rainfall anomalies as climate variables instead of a drought measure (Table A8).

Column 1 shows that (i) less rainfall than the long term average over the spring

season is associated with more disputes and (ii) that land certification reduces

significantly this effect on disputes. Column 2 shows that the effect of temper-

ature is less clear. While a deviation in temperature two periods ago increases

reported disputes, land certification does not seem to have any tempering effect.

Finally, column 3 reports the results of a placebo test: a negative water balance

out of the main rainy seasons (December, January and February) has no effect

on land disputes, as expected.

6 Land Certification, Water Scarcity and Land

Disputes: Underlying Mechanisms

Having identified the effect of water scarcity on land disputes and the dampening

effect of tenure security, we are now interested in studying possible underlying

mechanisms explaining these effects. Following our theoretical predictions pre-

sented in section 2, we investigate whether water scarcity increases the propensity

of land disputes through an increase in the marginal value of land (prediction 3),

and when there is more inequality in land distribution (prediction 4). In addi-

tion, we explore another plausible mechanism presented in the literature, that is

the role of access to credit in reducing land disputes through income smoothing.

6.1 The Marginal Value of Land

The prime objective is to shed light on the mechanisms linking water shocks,

property rights and disputes. First, we investigate whether climate stress has

a larger impact on land disputes when the marginal value of land is more im-

33

portant (prediction 3). We thus investigate whether having more labor available

on the farm (i.e. proxied by household size larger than the median) or having

more livestock than the median or whether the farm-household uses soil conser-

vation measures would increase the impact of water scarcity on land disputes.

We assume that having more labor or more livestock—crucial for farming and

herding—increases the marginal value of land (if thought as a traditional com-

plementary factor of production in an agricultural profit function). Conversely,

using soil conservation measures is expected to decrease the marginal value of

land by putting less pressure on each land unit.

Table 5 presents results when we include in equation (5) the factors affect-

ing the marginal value of land, the climate shock variables and the interaction

between them. Columns 1-2 present results when we include the interaction

terms SPEI Shocks × Household Size and SPEI Shocks × Livestock, re-

spectively. We observe a positive relationship between the interaction of actual

water scarcity with household size and livestock. This is supportive of prediction

3: water scarcity increases land disputes through an increase in the marginal

value of land. Column 3 includes the interaction term between SPEI Shocks

and a dummy variable indicating whether the farm-household adopted soil con-

servation measures. As expected, we find that farm-households are less prone to

land disputes in difficult times when adopting soil conservation measures.

In addition, in column 4 we study whether having more land available (i.e.

share of land xi in the theoretical model) decreases or increases the impact of wa-

ter scarcity on land disputes (prediction 4). We use a dummy variable indicating

if the farm size is larger than the median to investigate whether land inequal-

ity matters in difficult times, and we interact this dummy variable with SPEI

Shocks. We find that a farm-household having more land than the median (our

proxy for land inequality) will face more land disputes in times of water scarcity.

In summary, we consistently find that actual water availability has a stronger

34

Table 5: Land Certification, Water Scarcity and Land disputes: UnderlyingMechanisms

(1) (2) (3) (4) (5)Land disputes Land disputes Land disputes Land disputes Land disputes

Land certification -0.054 -0.054 -0.057 -0.047 -0.059(0.062) (0.060) (0.057) (0.063) (0.061)

Strong SPEI Shock (summer rainy season) -0.107 -0.173 0.111 -0.193 -0.064(0.093) (0.104) (0.119) (0.126) (0.082)

Strong SPEI Shock (spring rainy season) -0.017 -0.001 0.086 -0.026 0.054(0.067) (0.066) (0.101) (0.079) (0.074)

Household size -0.038(0.037)

Household size × Strong SPEI Shock (summer rainy season) 0.100**(0.036)

Household size × Strong SPEI Shock (spring rainy season) 0.081(0.098)

Livestock -0.054( 0.031)

Livestock × Strong SPEI Shock (summer rainy season) 0.220***(0.068)

Livestock × Strong SPEI Shock (spring rainy season) 0.062(0.064)

Adopted soil conservation 0.053(0.042)

Adopted soil conservation × Strong SPEI Shock (summer rainy season) -0.251*(0.126)

Adopted soil conservation × Strong SPEI Shock (spring rainy season) -0.102(0.123)

Farm size -0.080(0.037)

Farm size × Strong SPEI Shock (summer rainy season) 0.241**(0.084)

Farm size × Strong SPEI Shock (spring rainy season) 0.090(0.092)

Access to credit 0.033(0.046)

Access to credit× Strong SPEI Shock (summer rainy season) 0.067(0.115)

Access to credit× Strong SPEI Shock (spring rainy season) -0.142(0.135)

Village fixed effects yes yes yes yes yesYear fixed effects yes yes yes yes yesObservations 2734 2734 2734 2731 2731R2 0.023 0.025 0.026 0.024 0.024

Notes: All regressions are linear probability models. Variables are defined in Table 2 and Table 4. Standard

errors clustered at the kebele level are reported in parentheses. ∗ Significant at 10% level; ∗∗ Significant at the

5% level; ∗∗∗ Significant at the 1% level. 35

impact on the level of disputes when the marginal value of land is larger: either

because there is more labor or more livestock available in the farm, less conser-

vation measures or because there is more inequality in land distribution. These

findings are supportive of the predictions developed in the theoretical model pre-

sented in section 2: water scarcity increases the marginal value of land, which in

turns increases the temptation to rely on disputes instead of peaceful bargaining

in difficult times.

6.2 Alternative Mechanism: Access to Credit

Our theoretical framework predicts that water scarcity increases the propensity

of disputes through a rise in the marginal value of land. Property rights dampen

this effect by increasing the cost of encroaching. Another plausible mechanism

adduced in the literature highlights the role of income shocks in triggering dis-

putes: agents fight over a dwindling resource because of a strong and unexpected

drop in income. If well-defined property rights allow to smooth income over time,

they would also limit disputes. One way through which property rights may al-

low to smooth income is found in the literature on the ”De Soto” effect (Besley

et al., 2012): well-defined property rights can facilitate access to credit because

fixed assets can be used as collateral. While this would not invalidate the link

between land tenure and land disputes, it refers to a completely different mech-

anism. In particular, it would imply that income shocks and credit constraints

were the critical factors behind clashes.

We distinguish between these mechanisms by using the information available

on access to credit by each household. We investigate if weather stress has a lower

effect on land disputes when the household has access to credit. It is purported

that access to credit allows to smooth income and may thus dampen the effect of

weather shock. Column 5 of Table 5 presents the results including the interaction

between SPEI Shocks and Access to Credit. We find no significant evidence

36

supporting the idea that access to credit reduces the occurrence of land disputes

through income smoothing.

7 Concluding Remarks

Recent literature presents robust evidence that well-defined property rights over

land have a positive impact on many economic outcomes. In this paper, we

present a novel theoretical model of land disputes and use a large panel dataset

from Ethiopia to investigate the causal relationship between tenure security and

land disputes, and whether land certification could lessen the effects of water

scarcity on local clashes. Our identification strategy relies on the gradual rollout

of the land certification program in Ethiopia at the village level, the exogeneity

of climate shocks and the panel nature of our dataset that allows us to control

for village and time fixed effects.

First, we find that well defined property rights decrease the likelihood of

land disputes. Second, in line with the previous literature, we find that water

scarcity increases the likelihood of land disputes. Then, we highlight that land

certification decreases the effect of water scarcity on land disputes. Finally, we

show that actual water conditions have a stronger impact on land disputes when

the marginal value of land is larger: either because there is more labor or livestock

available in the farm or because there is more land inequality. Our findings are

robust to different specifications and measures of climate conditions.

These results suggest that land certification can be considered as an effective

policy instrument to buffer against water scarcity. The policy implications of

this paper are potentially very large. Policies that strengthen property rights

over land besides creating a precondition for economic growth and development

may also reduce the likelihood of disputes triggered by environmental shocks.

Secure property rights to land can have a profound effect on incentives and on

37

the working of markets for land and on welfare in general.

38

References

[1] Acemoglu, D. and I. N. Chaves and P. Osafo-Kwaako and J. A. Robinson

2014. Indirect Rule and State Weakness in Africa: Sierra Leone in Compara-

tive Perspective. African Economic Successes edited by Sebastian Edwards,

Simon Johnson and David Weil, University of Chicago Press, Chicago, IL..

[2] Adenew, B. and F. Abdi 2005. Research Report 3: Land Registration in

Amhara Region, Ethiopia. International Institute for Environment and De-

velopment.

[3] Alesina, A. and E. Spolaore. 2005. War, Peace and the Size of Nations.

Journal of Public Economics 89: 1333-1354.

[4] Almer C., J. Laurent-Lucchetti and M. Oechslin. 2015. Water Scarcity and

Rioting: Disaggregated Evidence from Sub-Saharan Africa. Mimeo.

[5] Andre, C., and J. P. Platteau. 1998. Land Relations under Unbearable

Stress: Rwanda Caught in the Malthusian Trap. Journal of Economic Be-

havior and Organization 34(1): 1-47.

[6] Begueria, S., and S. M. Vicente-Serrano. 2013. SPEI: Calculation of the

Standardised Precipitation-Evapotranspiration Index, (R package version

1.6 edn.).

[7] Besley, T., 1995. Property Rights and Investment Incentives: Theory and

Evidence from Ghana. Journal of Political Economy 103(5): 903-937.

[8] Besley, T. and M. Gathak. 2009. Property Rights and Economic Develop-

ment. In: Rodrick, Dani and Rosenzweig, M. R., (eds.) Handbook of Devel-

opment Economics. Handbooks in Economics, 5. Elsevier North Holland.

[9] Besley, T., K. Burchardi and M. Ghatak. 2012. Incentives and the De Soto

Effect. Quarterly Journal of Economics 127(1): 237-282.

39

[10] Bezabih, M., G. Kohlin, and A. Mannberg. 2011. Trust, Tenure Insecurity,

and Land Certification in Rural Ethiopia. Journal of Socio-Economics 40:

833-843.

[11] Binmore, K. and P. Dasgupta, eds. 1987. The Economics of Bargaining.

Oxford: Blackwell.

[12] Blattman C, Hartman A, Blair R. 2014. How to Promote Order and Prop-

erty Rights under Weak Rule of Law? An Experiment in Changing Dispute

Resolution Behavior Through Community Education. American Political

Science Review 108(1): 100-120.

[13] Blattman, C., and E. Miguel. 2010. Civil War. Journal of Economic Liter-

ature 48(1): 3-57.

[14] Cameron, C., J. Gelbach, and D. Miller. 2011. Robust Inference with Multi-

way Clustering. Journal of Business and Economic Statistics 29(2): 238-249.

[15] Chassang, S. and G. Padro-i-Miquel. 2009. Economic Shocks and Civil War.

Quarterly Journal of Political Science 4(3): 211-228.

[16] Couttenier, M., and R. Soubeyran. 2014. Drought and Civil War in Sub-

Saharan Africa. The Economic Journal 124(575): 201-244.

[17] Daly, C. 2006. Guidelines for Assessing the Suitability of Spatial Climate

Datasets. International Journal of Climatology 26(6): 707-721.

[18] Deaton, A. 1992. Understanding Consumption. Oxford University Press, Ox-

ford.

[19] Deininger, K., D. A. Ali, and T. Alemu. 2011. Impacts of Land Certification

on Tenure Security, Investment, and Land Market Participation: Evidence

from Ethiopia. Land Economics 8(2): 312-334.

40

[20] Deininger, K., D. A. Ali, S. Holden, and J. Zevenbergen. 2008. Rural Land

Certification in Ethiopia: Process, Initial Impact, and Implications for Other

African Countries. World Development 36(10): 1786-1812.

[21] de Janvry A., K. Emerick, M. Gonzalez-Navarro and E. Sadoulet. 2015.

Delinking Land Rights from Land Use: Certification and Migration in Mex-

ico. American Economic Review 105(10): 3125-3149.

[22] Fearon, J. D. 1998. Bargaining, Enforcement, and International Coopera-

tion. International Organization 52(2): 269-305.

[23] Fenske, J. 2011. Land Tenure and Investment Incentives: Evidence from

West Africa. Journal of Development Economics 95(2): 137-156.

[24] Fan, Y. and H. Van den Dool. 2008. A Global Monthly Land Surface Air

Temperature Analysis for 1948-present. Journal of Geophysical Research:

Atmospheres 113: D01103.

[25] Goldstein, M., and C. Udry. 2008. The Profits of Power: Land Rights and

Agricultural Investment in Ghana. Journal of Political Economy 116(6):

981-1022.

[26] Harari, M., and E. La Ferrara. 2014. Conflict, Climate and Cells: A Disag-

gregated Analysis. IGIER Working Paper n. 461.

[27] Hsiang, S.M. 2010. Temperatures and Cyclones Strongly Associated with

Economic Production in the Caribbean and Central America. Proc. Natl.

Acad. Sci. 107: 15367-15372.

[28] Jia, R. 2013. Weather Shocks, Sweet Potatoes and Peasant Revolts in His-

torical China. The Economic Journal 124(575): 92-118.

[29] Nickola, T. 1988. The Agricultural Sector in Ethiopia: Organization, Poli-

cies and Prospects in A. Penrose (ed.), Beyond the Famine: An Examination

41

of Issues behind the Famine in Ethiopia. Geneva: International Institute for

Relief and Development, Food for Hungry International.

[30] Novella, N. S., and W. M. Thiaw. 2013. African Rainfall Climatology Version

2 for Famine Early Warning Systems. Journal of Applied Meteorology and

Climatology 52(3): 588-606.

[31] Ostrom, E. 2005. Understanding Institutional Diversity. Princeton. Prince-

ton University Press.

[32] Palm, L. 2000. Quick and Cheap Mass Land Registration and Comput-

erisation in Ethiopia, Paper presented at FIG Congress 2010, ”Facing the

Challenges and Building the Capacity,” Sydney, Australia, 11-16 April 2010.

[33] Rubinstein, A. 1982. Perfect Equilibrium in a Bargaining Model. Economet-

rica 50(1): 97-109.

[34] United Nations Environment Programme 2011. Toolkit and Guidance for

Preventing and Managing Land and Natural Resources Conflict. Available

online at http://www.un.org/en/land-natural-resources-conflict/resource-

rich-economies.shtml (last accessed: March 09, 2016).

[35] Vicente-Serrano, S.M., S. Begueria, J. I. Lupez-Moreno, M. Angulo, A. El

Kenawy. 2010. A Global 0.5 Gridded Dataset (1901-2006) of a Multiscalar

Drought Index Considering the Joint Effects of Precipitation and Tempera-

ture. Journal of Hydrometeorology 11(4): 1033-1043.

[36] Wahba, G. 1990. Spline Models for Observational Data. Philadelphia: Soci-

ety for Industrial and Applied Mathematics.

42

A Appendix for Online Publication

A.1 Climate Indicators

The Standardised Precipitation-Evapotranspiration Index (SPEI) We

computed the Thornthwaite evapotranspiration index using the temperature

data and the latitude of the centroids of each kebele following the methodology

provided in Vicente-Serrano et al. (2010), and the code developed by Begueria

and Vicente-Serrano (2013). Evapotranspiration is then subtracted from precip-

itation in order to obtain the net balance of water in each geographical unit (the

centroid of each kebele). Finally, we standardized the net balance of water in

order to obtain the SPEI index (see Vicente-Serrano et al., 2010 for more details).

The climatic water balance is an important factor affecting vegetation ac-

tivity. According to Vicente-Serrano et al. (2010), the correlation between the

water balance and vegetation activity is particularly strong and immediate un-

der arid, semi-arid and sub-humid conditions, i.e. under conditions present in

many parts of Ethiopia’s agricultural regions. This, together with the strong

correlation between SPEI and vegetation activity, suggests that the actual water

balance matters for agricultural productivity and other types of water-dependent

economic activities, like pastoralism.

Our benchmark indicator of climate shock, denoted as SPEI Shock, captures

low SPEI episodes occurring during the main rainy season (the ”Meher”: June,

July, August and September) and during the secondary rainy season (the ”Belg”)

happening in Spring with a construction of the same variable over March, April

and May. Our indication of climate shocks is defined as follows: for 2007, we

consider the four months of the rainy season from the last two years (2006 and

2005) and take the number of months in which SPEI was below its mean by some

magnitude. We use two thresholds: a SPEI between −0.5 and −1 and below −1.

Then, divide this number by eight, the total number of months of the Meher over

43

two years.26 The value of our main SPEI variable thus ranges between 0 and 1,

with 0 denoting a very good period in which SPEI never went below its mean

over the last two years during the rainy season, and 1 denoting an exceptionally

”bad” period in which the rainy season of the last two years was plagued with

abnormally low values of SPEI.

We also use as a robustness check rainfall and temperature data at the house-

hold level. We impute the household specific rainfall and temperature values us-

ing longitude, latitude and elevation information of each household by adopting

the Thin Plate Spline method of spatial interpolation, which is commonly used

to generate spatial climate datasets.27 This method has the advantage that it

accounts for spatially varying elevation relationships and it is not difficult to ap-

ply. However, it does not handle easily very sharp spatial gradients, and it only

simulates elevation relationship. This is a typical characteristic of coastal areas.

Significant terrain features, and no climatically important coastlines characterise

our study area (for more details on the properties of this method see Daly, 2006).

We finally use rainfall anomalies as a measure of climate anomalies, that is the

difference between the weather at time t and the 1976-2006 climatic data divided

by the 1976-2006 standard deviation.

A.2 Figures

26Here we follow Harari and La Ferrara (2014) to construct a climate indicator including an”intensity” component and accounting for two years lags (previous weather conditions).

27By definition, Thin Plate Spline is a physically based two-dimensional interpolation schemefor arbitrarily spaced tabulated data. The Spline surface represents a thin metal sheet that isconstrained not to move at the grid points, which ensures that the generated rainfall data atthe weather stations are exactly the same as the data at the weather station sites that wereused for the interpolation. In our case, the rainfall data at the weather stations are reproducedby the interpolation for those stations, which ensures the credibility of the method (see Wahba,1990 for details).

44

Figure A1: Map of the Amhara Region, Ethiopia (Source: Deininger et al., 2011)

45

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1999 2002 2005 2007

Pro

po

rtio

n o

f H

ou

seh

old

se

xp

ect

ing

a

cha

ng

e i

n l

an

d a

llo

cati

on

in

th

e n

ex

t5

ye

ars

Perception of tenure insecurity

Villages with Tenure

Villages without Tenure

Tenure certification

(treatment)

Land redistribution

program (1997)

Figure A2: Expectations of Land Redistribution in the Next Five Years

46

Figure A3: Water Scarcity, Probability of Disputes and Tenure Security

47

A.3 Descriptive Statistics at the Village and Household

Level

Table A1: Descriptive Statistics at the Village Level

Full sample Non-certified Certified Non-certifiedMean Std. Dev. Mean Std. Dev. Mean Std. Dev. vs certified

Panel A: Year 2005Land disputes 0.189 0.037 0.192 0.03 0.183 0.055 n.s.Male household head 0.820 0.044 0.824 0.032 0.811 0.067 n.s.Age of household head 50.315 2.474 50.62 3.025 49.705 0.589 n.s.Literate household head 0.475 0.078 0.446 0.075 0.534 0.047 *Household size 6.36 0.583 6.329 0.511 6.420 0.793 n.s.Farm size (ha) 1.821 0.576 1.766 0.355 1.932 0.947 n.s.Total livestock 4.258 1.150 3.87 0.909 5.035 1.309 *Access to credit 0.182 0.092 0.200 0.094 0.148 0.089 n.s.

Panel B : Year 2007Land disputes 0.226 0.108 0.26 0.104 0.159 0.095 n.s.Male household head 0.817 0.051 0.823 0.039 0.804 0.075 n.s.Age of household head 51.519 2.345 51.997 2.735 50.562 0.939 n.s.Literate household head 0.432 0.079 0.391 0.059 0.513 0.037 ***Household size 6.667 0.642 6.604 0.551 6.793 0.877 n.s.Farm size (ha) 2.082 0.847 1.901 0.3 2.443 1.469 n.s.Total livestock 4.462 1.326 4.053 0.995 5.28 1.673 n.s.Access to credit 0.240 0.075 0.225 0.074 0.269 0.079 n.s.Observations 12 8 4

Notes: n.s. indicates not significant difference between non-certified and certified villages. ∗ Significant at 10%

level; ∗∗∗ Significant at the 1% level.

48

Table A2: Descriptive Statistics at the Farm-household Level

Full sample Non-certified Certified Non-certifiedObs. Mean Std. Dev. Obs. Mean Std. Dev. Obs. Mean Std. Dev. vs certified

Panel A: Year 2005Land disputes 1,487 0.189 0.392 1,027 0.192 0.394 460 0.183 0.387 n.sMale household head 1,486 0.82 0.385 1,027 0.824 0.381 459 0.81 0.392 n.sAge of household head 1,485 50.562 15.43 1,027 50.945 15.631 458 49.703 14.95 n.s.Literate household head 1,441 0.477 0.5 1,001 0.452 0.498 440 0.534 0.499 ***Household size 1,438 6.351 2.356 997 6.316 2.358 441 6.431 2.353 n.s.Farm size (ha) 1,438 1.814 3.092 997 1.755 3.173 441 1.945 2.899 n.sTotal livestock 1,438 4.205 3.427 997 3.831 3.051 441 5.053 4.03 ***Access to credit 1,487 0.18 0.385 1,027 0.195 0.396 460 0.148 0.355 **

Panel B : Year 2007Land disputes 1,487 0.221 0.415 1,027 0.248 0.432 460 0.159 0.366 ***Male household head 1,486 0.817 0.387 1,027 0.823 0.382 459 0.804 0.397 **Age of household head 1,485 51.684 15.009 1,027 52.186 15.31 458 50.557 14.264 **Literate household head 1,441 0.432 0.495 1,001 0.396 0.489 440 0.514 0.5 ***Household size 1,438 6.649 2.442 997 6.58 2.409 441 6.805 2.512 n.s.Farm size (ha) 1,438 2.057 1.869 997 1.878 1.636 441 2.463 2.26 ***Total livestock 1,438 4.398 3.615 997 3.998 3.221 441 5.303 4.245 ***Access to credit 1,487 0.235 0.424 1,027 0.22 0.414 460 0.27 0.444 ***

Notes: n.s. indicates not significant difference between non-certified and certified farm-households. ∗∗ Signifi-

cant at the 5% level. ∗∗∗ Significant at the 1% level.

49

A.4 Land Certification and Land Disputes: Propensity

Score Matching

Table A3: Propensity Score Matching: Land Disputes in 2005

(1) (2) (3) (4)Land disputes: certified Land disputes: Non-certified Differences Standard errors

Average unmatched 0.178 0.196 -0.018 0.023Average matched 0.178 0.147 0.031 0.032

Observations 419 956

Table A4: Propensity Score Matching: Land Disputes in 2007

(1) (2) (3) (4)Land disputes: certified Land disputes: Non-certified Differences Standard errors

Average unmatched 0.162 0.249 -0.087∗∗∗ 0.024Average matched 0.162 0.265 -0.103∗∗∗ 0.035

Observations 407 907

Notes: ∗∗∗ Significant at the 1% level.

50

A.5 Additional Robustness Analysis

Table A5: Baseline Results: Different Clustering Methods

(1) (2) (3)Land disputes Land disputes Land disputes

Land certification -0.080∗∗ -0.080∗∗ -0.089∗

(0.042) (0.033) (0.048)

Village fixed effects yes yes noYear fixed effects yes yes yesClustering Kebele (small sample) Kebele-Year Spatial correlationN 2974 2974 2974R2 0.027 0.037 0.123

Notes: Standard errors in parentheses. Column 1 presents standard errors with finite sample correction

(Cameron et al., 2011). Column 2 presents two-way cluster-robust standard errors. Column 3 presents standard

errors adjusted for spatial autocorrelation (Hsiang, 2010). ∗ Significant at 10% level; ∗∗ Significant at the 5%

level.

Table A6: Non-linear Specifications: Probit and Logit Models

(1) (2)Land disputes Land disputes

Land certification -0.294∗ -0.510∗

(0.159) (0.276)

Village fixed effects yes yesYear fixed effects yes yesObservations 2974 2974

Notes: Standard errors clustered at the kebele level are reported in parentheses. ∗ Significant at 10% level.

51

Table A7: Alternative Measures of Land Disputes

(1) (2)Onset of land disputes Decrease of land disputes

Land certification -0.029∗∗∗ 0.034∗∗∗

(0.000) (0.000)

Village fixed effects yes yesYear fixed effects yes yesObservations 1487 1487

Notes: Standard errors clustered at the kebele level are reported in parentheses. ∗∗∗ Significant at the 1% level.

52

Table A8: Alternative Measures of Water Scarcity: Rainfall and TemperatureAnomalies

(1) (2) (3)Land disputes Land disputes Land disputes

Land certification -0.095 -0.075∗ -0.138(0.056) (0.040) (0.096)

Deviation in rain (spring season, t-1) -0.056∗

(0.030)

Deviation in rain (spring season, t-1)×Land certification 0.323∗∗

(0.121)

Deviation in rain (spring season, t-2) -0.005(0.053)

Deviation in rain (spring season, t-2)×Land certification -0.025(0.064)

Deviation in temperature (t-1) -0.014(0.060)

Deviation in temperature (t-1) ×Land certification -0.020(0.196)

Deviation in temperature (t-2) 0.012∗∗

(0.005)

Deviation in temperature (t-2) ×Land certification -0.014(0.009)

Placebo: SPEI (off any rainy season, t-1) 0.032(0.025)

Placebo: SPEI (off any rainy season t-1)×Land certification 0.065(0.140)

Placebo: SPEI (off any rainy season t-1) 0.004(0.019)

Placebo: SPEI (off any rainy season t-1)×Land certification -0.043(0.092)

Village fixed effects yes yes yesYear fixed effects yes yes yesObservations 2723 2974 2734R2 0.029 0.028 0.023

Notes: Standard errors clustered at the kebele level are reported in parentheses. ∗ Significant at 10% level; ∗∗

Significant at the 5% level.

53